Action Observation Combined With Virtual Reality Promotes Motor Recovery After Stroke: A Randomized Controlled Trial

Clinical Question (PICO)

Population: Adults 3–18 months after stroke with unilateral upper-limb paresis and residual movement ability, receiving inpatient rehabilitation in two Italian centers.

Intervention: Action observation followed by imitation-based virtual-reality upper-limb training (AO+VR), 20 sessions over 5 weeks.

Comparison: The same dose and virtual-reality exercises, preceded by neutral natural-scene videos rather than observation of goal-directed actions (CO+VR).

Outcomes: Primary — change in paretic-hand dexterity on the Box and Block Test (BBT) at treatment end and 6-month follow-up. Secondary — nonparetic-hand dexterity, strength, spasticity, global disability, and functional independence. The authors explored whether age, time from stroke, baseline hand use, and cognitive status modified treatment response.

Bottom Line

In this small assessor-blinded randomized trial, adding action observation to virtual-reality rehabilitation produced greater paretic-hand dexterity improvement than virtual reality alone: the between-group difference in change was 7.8 BBT blocks at treatment end (95% CI, 7.1–7.9) and 10.8 blocks at 6 months (95% CI, 10.6–10.9). The AO+VR group’s mean gain was 8.0 blocks, above the prespecified 7-block smallest-detectable-difference benchmark, while strength, spasticity, disability, and activities of daily living improved similarly in both groups.

Design

Trial type: Multicenter, assessor-blinded, 1:1 randomized controlled trial; participants and treating clinicians were not blinded.
N: 48 randomized
- AO+VR: 24
- CO+VR: 24
Randomization: Computer-generated 1:1 allocation with concealed permuted blocks of varying size, stratified by clinical level of hand use (minimal vs discrete use).
Setting: Two inpatient stroke-rehabilitation centers in Italy: Centro Cardinal Ferrari in Parma and Clinical Institute Quarenghi in Bergamo.
Enrollment: January 2022–September 2024.
Mean follow-up: 6 months after treatment completion, with posttreatment testing within 72 hours.
Analysis: Intention-to-treat generalized estimating-equation models for repeated measures, with time, treatment, and time×treatment terms; Sidak-adjusted pairwise comparisons. Missing T1/T2 data were handled by treatment-group mean imputation. Exploratory models assessed covariate interactions.
Primary outcome: Change in unilateral BBT performance for the paretic hand; the BBT counts 25-mm blocks transferred in 60 seconds. The study used 7 blocks as the smallest detectable difference because no formal stroke-specific minimal clinically important difference was established.

Population

Inclusion Criteria

Adult with a clinical diagnosis of stroke 3–18 months after the acute event.
Primarily motor sequelae with unilateral upper-limb paresis and residual movement ability.
Inpatient admission to one of the participating rehabilitation centers.

Exclusion Criteria

Severe motor impairment without sufficient residual movement for the protocol.
Severe neuropsychological deficits that prevented participation.

Baseline Characteristics (overall or representative arm)

The groups were reported as similar in age, sex, time from stroke, affected side, hand-use level, and treatment duration; baseline BBT, Motricity Index, modified Ashworth Scale, modified Rankin Scale, and modified Barthel Index also did not differ statistically.
Three control participants withdrew before follow-up because of worsening clinical conditions, accidental trauma, or unrelated medical treatment; no AO+VR participant withdrew.

Interventions

AO+VR: Twenty 45-minute sessions, four times weekly for 5 weeks. Each session began with approximately 1.5 minutes of silent videos showing goal-directed unilateral or bilateral upper-limb actions, followed immediately by imitation in a semi-immersive VR system. Participants manipulated sensor-equipped physical objects while completing personalized exercises; the protocol included 52 exercises with therapist-adjusted facilitation.
CO+VR: Twenty sessions with the same frequency, duration, VR system, therapist support, and exercise dose. The observation phase used approximately 1.5 minutes of neutral natural-scene video without motor content, followed by the same VR exercises.
Both groups also received the centers’ usual multidisciplinary inpatient rehabilitation, including physiotherapy, occupational therapy, and task-oriented functional training.

Outcomes

Primary Outcome (between-group change in paretic-hand BBT score):

Paretic-hand BBT improved in both groups, but the time×group interaction favored AO+VR (Wald χ² = 3263.1; P < 0.001).
Mean change from baseline to treatment end was +8.0 blocks (SD 5.1) with AO+VR versus +2.6 blocks (SD 4.0) with CO+VR. The between-group difference in change was 7.8 blocks (95% CI, 7.1–7.9); at 6 months it was 10.8 blocks (95% CI, 10.6–10.9). These were prespecified primary time points, not post-hoc endpoints.
The AO+VR mean gain exceeded the study’s 7-block smallest-detectable-difference benchmark; the control-group mean gain did not. This benchmark is not the same as a formally established minimal clinically important difference.
Nonparetic-hand BBT also improved over time in both groups (Wald χ² = 21.1; P < 0.001), with comparable improvement between treatments.

Secondary Outcomes:

Motricity Index improved over time in both groups (Wald χ² = 73.5; P < 0.001), without a significant treatment×time interaction.
Modified Ashworth Scale scores decreased over time (Wald χ² = 35.45; P < 0.001), without a treatment-specific effect.
Modified Rankin Scale improved over time (Wald χ² = 74.5; P < 0.001), without a significant group or interaction effect.
Modified Barthel Index improved over time (Wald χ² = 23.6; P < 0.001), again without a significant between-group difference.
The authors described 100% completion of the assigned intervention program among participants who remained in follow-up and no detected adverse event.
Exploratory moderation analysis found an age×time-from-stroke interaction for paretic-hand BBT improvement (Wald χ² = 16.1; P < 0.001), with larger gains among younger participants and those closer to stroke onset. Treatment×covariate interactions for baseline hand use, age, and cognitive status were otherwise not significant; these analyses were exploratory.

Adverse Events / Safety:

No adverse event was attributed to either rehabilitation protocol. Three control-group participants withdrew because of clinical deterioration or unrelated medical care; the report did not identify these as intervention-related harms.

Figures

CONSORT flow diagram for the randomized trial — *Figure 1.* Flow diagram of the clinical trial according to CONSORT guidelines. All randomized participants were included in the intention-to-treat analysis; missing posttreatment or follow-up data were handled by mean imputation.

Experimental setup and action-observation protocol — *Figure 2.* Experimental setup and study protocol for AO+VR and CO+VR, including the action-observation video and the neutral natural-scene control video.

Mean change in primary and secondary outcomes over time — *Figure 3.* Mean change over time in the primary and secondary outcome measures for AO+VR and CO+VR at baseline, treatment end, and 6-month follow-up; error bars represent mean standard error.

Age and time-from-stroke interaction for dexterity improvement — *Figure 4.* Interaction between age, time from stroke onset, and paretic-hand BBT improvement from generalized estimating-equation analysis; error bars represent mean standard error.

Criticisms

The trial randomized only 48 participants, although the prespecified calculation called for 94; the resulting estimates are unusually precise in the abstract’s reported confidence intervals and need replication in a larger sample.
The primary measure was the BBT, and the paper did not include core upper-limb measures such as the Fugl-Meyer Assessment or Action Research Arm Test. A dexterity signal therefore does not establish broad upper-limb recovery or better real-world arm use.
Participants and treating clinicians knew the assigned treatment. The attention, expectancy, and novelty of watching goal-directed actions may contribute to the observed effect even though session dose and VR exercises were matched.
Three control participants were lost before follow-up, and missing values were replaced with the treatment-group mean. Mean imputation can narrow uncertainty and distort repeated-measures estimates, particularly in a small trial.
The age and time-from-stroke interaction was exploratory. It should not be used as a validated rule to preferentially offer AO+VR to younger or earlier presenters.
Both groups received substantial usual inpatient rehabilitation, and the study was performed in two experienced VR centers. External validity to outpatient programs, resource-limited settings, severe paresis, or patients unable to engage with VR is uncertain.

Funding

Funded by a grant from the Italian Ministry of Health to Dr Errante (SG-2019-12370506). The grant proposal underwent peer review, but the Ministry had no role in study design or article writing. The trial was registered at ClinicalTrials.gov (NCT05163210). The authors reported no disclosures.

The paper

Authors. Errante et al.
Title. Action Observation Combined With Virtual Reality Promotes Motor Recovery After Stroke: A Randomized Controlled Trial.
Journal. Stroke.
Year. 2026.
DOI. 10.1161/STROKEAHA.125.054101
PMCID. PMC13117561

Deep Dive — click to expand

What this is

This is a small randomized test of whether the content of the pre-exercise visual stimulus matters when the rehabilitation dose is otherwise held constant. Forty-eight inpatients 3–18 months after stroke received the same semi-immersive VR motor training; only the AO+VR group first watched goal-directed upper-limb actions and then imitated them. The headline result is a paretic-hand BBT advantage that exceeded the study’s 7-block reliability benchmark at treatment end and was larger at 6 months, while broader strength, tone, disability, and independence measures did not separate between groups.

1. Shadow Audit

The framing risks making the study sound like a broad motor-recovery trial when its clearest between-group signal is a single dexterity test. Both groups improved in Motricity Index, spasticity, modified Rankin Scale, and modified Barthel Index, but none of those measures showed a treatment-specific interaction. The authors’ clinically useful distinction is therefore narrow: AO+VR may add a paretic-hand dexterity component to VR, not necessarily extra independence, strength, or global disability recovery.

The most important methodological shadow is the missing-data strategy. Three of 24 control participants did not complete follow-up, and the trial replaced missing values with the control-group mean at the relevant time point. In a 48-person study, that can make the trajectory look more orderly than the patients’ actual outcomes and can understate uncertainty. The abstract’s dramatic 7.8- and 10.8-block differences should be read alongside the modest, under-recruited sample and the absence of a formal BBT minimal clinically important difference.

2. Inversion Engine

For the practical conclusion to invert, a larger, independently run trial would need to show that the BBT separation is primarily expectancy, attention, or measurement behavior rather than a reproducible motor effect. The study itself supplies a useful threshold: AO+VR’s mean treatment-end gain was 8.0 blocks, just above the 7-block smallest-detectable-difference benchmark, while CO+VR gained 2.6. An inversion would occur if that roughly 5.4-block within-trial contrast shrank below measurement error or disappeared when missing data were handled with multiple imputation or a mixed model, and when outcomes included Fugl-Meyer, Action Research Arm Test, and patient-reported arm use.

The long-term claim is also conditional. The between-group difference was reported as 10.8 blocks at 6 months, but only 21 of 24 control participants completed follow-up. If the three missing control outcomes were systematically worse or better than the imputed mean, the magnitude and precision of the persistence claim could change substantially.

3. Second-Order Catalyst

If the result is real, the first adopters are not acute stroke units or emergency stroke teams; they are inpatient and outpatient neurorehabilitation services that already own a VR platform and can script short action-observation clips. The lowest-friction protocol update is a five-week module: before each matched VR exercise, show a brief lateral-view recording of the actual goal-directed upper-limb action, then have the patient imitate it immediately. Therapists should personalize the task to residual movement and record BBT plus a functional arm measure at baseline, discharge, and 6 months.

Implementation should start as a measured service improvement, not a wholesale replacement of conventional therapy. A center could pilot the protocol in patients able to attend 3–4 sessions per week, track completion and adverse events, and require a functional measure beyond BBT before claiming meaningful benefit.

4. Asymmetric Leverage

The payoff is asymmetric because the added ingredient is small—roughly 1.5 minutes of goal-directed observation before the same VR task—but it targets a large rehabilitation denominator: every eligible session in a service that already uses VR. If that brief cue reliably produces even a modest dexterity gain without extra equipment or treatment time, it could scale across many sessions and patients.

The counterweight is that the leverage is only large if the effect survives outside two highly experienced Italian centers. The study’s secondary outcomes did not show a between-group advantage, so the high-upside claim is an inexpensive protocol refinement, not evidence that a costly VR platform should be purchased solely for this intervention.

5. Paradigm Destroyer

The reflex this paper challenges is treating VR rehabilitation as a generic motor-repetition dose in which the display content is interchangeable. Here, the same VR exercise was paired with either goal-directed action observation or neutral nature scenes, and the paretic-hand dexterity signal favored the action-linked preparation.

Protocol update for tomorrow morning: For patients with residual upper-limb movement who are already doing VR exercises, precede each task with a short, task-matched video of the intended action and immediate imitation. Do not infer global recovery from BBT alone; pair the change with an arm-specific functional measure and document whether the patient can transfer the task to daily activity.

MVP — Minimum Viable Proof

The minimum practice-changing statement is: in a larger pragmatic trial with robust missing-data handling, adding brief task-matched action observation to the same VR dose produces a reproducible paretic-hand BBT advantage of at least 7 blocks and also improves an independent functional arm measure without increasing treatment time or adverse events.

Best Combination

The most defensible synthesis is to combine this trial with the broader action-observation and VR rehabilitation literature as a component test: action observation is biologically plausible and inexpensive, while VR supplies adjustable, repeatable task practice. This trial supports adding the observation component to an existing VR program, but it does not establish that AO+VR is superior to conventional task-oriented therapy, nor that it improves activities of daily living beyond the improvement produced by VR plus usual rehabilitation.

The next evidence layer should be a larger, adequately powered, multicenter trial using blinded assessment, multiple imputation or mixed-model sensitivity analyses, and standardized arm outcomes such as Fugl-Meyer and Action Research Arm Test alongside BBT. That design would distinguish a durable, transferable motor benefit from a task-specific performance advantage.

Overvalue Warning

Do not convert the 7-block smallest-detectable-difference benchmark into a proven minimal clinically important difference; the paper explicitly notes that a formal stroke-specific MCID for BBT has not been established.
Do not treat the exploratory finding that younger patients and those closer to stroke onset improved more as a validated selection rule. The trial was small, the interaction was exploratory, and meaningful benefit may still occur later after stroke.

Educational use notice. The Clotbust Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.